MODULI OF PARABOLIC HIGGS BUNDLES AND
ATIYAH ALGEBROIDS
MARINA LOGARES AND JOHAN MARTENS
Abstract. In this paper we study the geometry of the moduli
space of (non-strongly) parabolic Higgs bundles over a Riemann
surface with marked points. We show that this space possesses a
Poisson structure, extending the one over the dual of an Atiyah algebroid over the moduli space of parabolic vector bundles. By considering the case of full flags, we get a Grothendieck-Springer resolution for all other flag types, in particular for the moduli spaces of
twisted Higgs bundles, as studied by Markman and Bottacin and
used in the recent work of Laumon-Ngô. We discuss the Hitchin
system, and demonstrate that all these moduli spaces are integrable systems in the Poisson sense.
Contents
1. Introduction
1.1. Remark on notation
1.2. Acknowledgements
2. Moduli spaces of Parabolic Higgs bundles
2.1. Parabolic vector bundles
2.2. Parabolic Higgs bundles
2.3. The Hitchin fibration
2.3.1. Hitchin map
2.3.2. Spectral curves
2.3.3. Generic fibers
3. Poisson structure
3.1. Background material
3.1.1. Poisson geometry
3.1.2. Algebroids and Poisson structures
3.1.3. Symplectic leaves for the dual of an algebroid
3.2. Poisson structure on Pα
3.2.1. The ] map
3.2.2. Poisson structure via Lie algebroids
3.2.3. Extension of the bracket
Date: June 2, 2009.
1
2
4
4
4
4
8
8
9
9
10
12
12
12
13
15
16
16
17
19
2
MARINA LOGARES AND JOHAN MARTENS
3.2.4. Symplectic leaves
3.2.5. Complete integrability of the Hitchin system
4. Morphisms between moduli spaces, Grothendieck-Springer
resolution
5. Comparison with Bottacin-Markman and further remarks
Appendix A. A Levi principal bundle over the moduli space of
parabolic bundles
References
20
22
24
26
28
29
1. Introduction
Higgs bundles, introduced by Hitchin in [Hit87a, Hit87b], have emerged in the last two decades as a central object of study in geometry,
with several links to physics and number theory. Over a smooth compact Riemann surface the moduli space of Higgs bundles contains as
a dense open subset the total space of the cotangent bundle to the
moduli space of vector bundles. In fact the induced complex symplectic form is part of a hyper-Kähler structure and extends to the whole
of the moduli space of Higgs bundles, and it is a celebrated fact that
the moduli space comes equipped with a algebraically completely integrable system, through the Hitchin map.
A natural generalization of vector bundles arises when one endows
the vector bundle with a parabolic structure [MS80], i.e. with choices
of flags in the fibers over certain marked points on the Riemann surface.
One can talk of Higgs bundles in that setting as well, as was first done
by Simpson in [Sim90]. Various choices can be made for this. In order to have the corresponding moduli space contain as an open subset
the total space of the cotangent bundle to the moduli space of parabolic vector bundles, replicating the non-parabolic situation, several
authors [Tha02, Kon93, GPGM07, Fal93] restrict the parabolic Higgs
bundles to those that we shall refer to as strongly parabolic, meaning
that the Higgs field is nilpotent with respect to the flag.
One can however also demand the Higgs field to simply respect the
parabolic structure at the marked points, and a moduli space Pα for
those was constructed by Yokogawa in [Yok93]. The locus of that
moduli space where the underlying parabolic vector bundle is stable
again forms a vector bundle over the moduli space of parabolic vector bundles Nα . We show here that this vector bundle is the dual
of an Atiyah algebroid associated with a principal bundle over Nα ,
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS
3
the structure group for which is the product of the Levi groups given
by the various flags at the marked points, modulo C∗ to account for
global endomorphisms of the bundle. As the dual of an algebroid its
total space carries a complex algebraic Poisson structure, which in fact
extends to the whole of Pα . In the particular situation where all the
flags are trivial, {0} ⊂ E|p , this was already shown independently by
Bottacin [Bot95] and Markman [Mar94]. We further study the Hitchin
system for Pα and its symplectic leaves, showing that this makes the
Pα for all flag types into integrable systems in the sense of Poisson geometry, with Casimir functions that generically induce the foliation of
symplectic leaves, wich are integrable systems in the usual symplectic
sense. Though we don’t explicitly use Lie groupoids, our philosophy is
very much that the symplectic leaves are the co-adjoint orbits for the
groupoid determined by the principal bundle.
With this in mind we also look at the forgetful morphisms between
such moduli spaces of various flag types. We show that they are Poisson and generically finite. By looking at such morphisms starting from
the moduli space for full flags we obtain a global analogue of the Grothendieck-Springer resolution of Lie algebras, as the moduli space for
full flags is a regular Poisson manifold. As the Grothendieck-Springer
resolution plays a crucial role in modern geometric representation theory, this opens perspectives on generalizing classical constructions to
this global setting1.
The presentation we have given is largely done in the language of Lie
algebroids, but one could have reformulated everything we say about
Atiyah algebroids in terms of Poisson reduction of cotangent bundles.
The choice is partly one of personal preference, and partly due to the
fact that the Atiyah sequence of the algebroid naturally follows from
the deformation theory of parabolic vector bundles. Similarly, though
almost all of our arguments use hypercohomology, we have avoided the
use of derived categories, in order to exhibit the Poisson structures
more explicitly.
We remark here that our entire construction is depending on the
existence of the principal bundle over Nα . In the appendix A we describe a construction of this bundle for full flags, using previous work
1Indeed,
we very recently became aware of the preprint [Yun08], where parabolic Higgs bundles are used to generalize the Springer theory of Weyl group representations to a ‘global’ setting, without however taking our viewpoint of Atiyah
algebroids and groupoids.
4
MARINA LOGARES AND JOHAN MARTENS
by Hurtubise, Jeffrey and Sjamaar [HJS05], and outline a possible construction for other flag types.
This paper is organized as follows: in section 2 we give the necessary
background regarding parabolic Higgs bundles and their moduli, as well
as a description of the Hitchin fibration. Most of this is standard, with
the possible exception of the observation (Proposition 2.2) that the
smoothness of the spectral curve implies that the Higgs field uniquely
determines the parabolic structure, even when eigenvalues are repeated.
In section 3 we give the required background material regarding Lie
groupoids and Lie algebroids, and prove the main result of this article,
the interpretation of the moduli space as a partial compactification of
the dual of an Atiyah algebroid. We also show here that for the induced
Poisson structure the Hitchin map is an integrable system. In section 4
we remark that for nearby parabolic weights the morphisms between
the various moduli spaces (with different flag structures) are Poisson,
giving a Grothendieck-Springer resolution by means of the full flags. In
section 5 we discuss the relationship of our work with the earlier results
by Bottacin and Markman, as well as further directions. Appendix A
discusses a construction of the principal bundle over the moduli space
of parabolic vector bundles used in the main theorem.
1.1. Remark on notation. Unfortunately nomenclature conventions
regarding parabolic Higgs bundles vary in the literature. For us a
parabolic Higgs bundles will only require the Higgs field at a marked
point to respect the flag there. We will refer to the special case where
the Higgs field is nilpotent with respect to the filtration as a strongly
parabolic Higgs bundle.
1.2. Acknowledgements. The authors would like to thank Sergey
Arkhipov, David Ben-Zvi, Philip Boalch, Hans Boden, Chris Brav, Ron
Donagi, Tomás Goméz, Peter Gothen, Tamás Hausel, Nigel Hitchin,
Jacques Hurtubise, Lisa Jeffrey, Raj Mehta, Eckhard Meinrenken, Szilárd Szabó and Michael Thaddeus for useful conversations, remarks,
and encouragement, as well as the MPI Bonn, NSERC and the Centro
de Matemática da Universidade do Porto for financial support.
2. Moduli spaces of Parabolic Higgs bundles
2.1. Parabolic vector bundles. Let X be a compact Riemann surface or smooth complex projective curve of genus g with n distinct
marked points p1 , . . . , pn . Let D be the effective reduced divisor p1 +
. . . + pn . A parabolic vector bundle ([MS80]) on X is an algebraic rank
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS
5
r vector bundle E over X together with a parabolic structure, i.e. a
(not necessarily full) flag for the fiber of E over the marked points
E|p = Ep,1 ⊃ · · · ⊃ Ep,r(p) ⊃ 0,
together with a set of parabolic weights
0 ≤ α1 (p) < · · · < αr(p) (p) < 1.
We denote the multiplicities by mi (p) = dim Ep,i − dim Ep,i+1 , and the
associated graded as Gr(p) = ⊕i Ep,i /Ep,i+1 .
Note that the structure group of the bundle E is GL (r). In terms
of the associated frame bundle the parabolic structure corresponds to
a reduction of the structure group of this principal bundle to a certain
parabolic subgroup of GL (r) at each marked point pi . We will denote
this parabolic subgroup by Pp , and its corresponding Levi-group by Lp ,
with Lie algebra lp . For the sake of convention we will fix as a Borel
subgroup in GL(r) the lower triangular matrices, and all parabolic subgroups are taken to contain this Borel.
We will further need linear endomorphisms2 Φ of a parabolic vector
bundle which are either parabolic - meaning that at the fiber over
a marked point p we have Φ|p (Ep,i ) ⊂ Ep,i - or strongly parabolic
- meaning that Φ|p (Ep,i ) ⊂ Ep,i+1 . We denote the sheaves of parabolic respectively strongly parabolic endomorphisms as ParEnd(E)
and SParEnd(E).
The relevance of the parabolic weights α comes from the notion of
parabolic degree of a bundle, denoted by pdeg :
XX
pdeg(E) = deg(E) +
mi (p)αi (p),
p∈D
i
which satisfies the Gauss-Chern formula for connections with logarithmic singularities (see Proposition 2.9 in [Biq91]). The α also occur
in the celebrated Mehta-Sheshadri theorem ([MS80]) that establishes
a correspondence between stable parabolic bundles and unitary representations of the fundamental group of the punctured surface X \ D,
where they determine the holonomy around the punctures.
Every algebraic subbundle F of E is naturally given the structure of a
parabolic bundle as well, by simply intersecting F |p with the elements of
the flag of E|p , discarding any such subspace of F |p that coincides with
2A
rank.
priori we don’t require morphisms between vector bundles to have constant
6
MARINA LOGARES AND JOHAN MARTENS
a previous one, and endowing it with the largest of the corresponding
parabolic weights:
for Fp,i = F |p ∩ Ep,j
αiF (p) = max{αj ; F |p ∩ Ep,j = Fp,i }.
j
We say that a parabolic vector bundle is stable if for each proper
subbundle F we have that
pdeg(F )
pdeg(E)
(1)
<
.
rk (F )
rk (E)
Semistability is defined similarly, by asking for weak inequality. The
weights are called generic when stability and semistability coincide.
Note that the term generic is used in the sense that the set of nongeneric weights has positive codimension.
There exists a moduli space for parabolic vector bundles [MS80],
which we shall denote by Nα . This is a normal projective variety of
dimension
!
X1
X
2
2
2
dim Nα = (g − 1)r + 1 +
r −
mi,p
2
i
p∈D
and when the weights are generic it is smooth. From now one we will
assume genericity of weights, even though in the non-generic case all of
what we say can still be carried through when restricted to the stable
locus of Nα .
We will need a further lemma that states that stability implies simplicity. This is completely standard in the non-parabolic setting (see
e.g. [NS65, Corollary to Proposition 4.3]) but a parabolic version does
not seem to have appeared in the literature, so we include it here for
completeness.
Lemma 2.1. If E is an α-stable parabolic vector bundle over X, then
H 0 (X, ParEnd(E)) = C
and
H 0 (X, SParEnd(E)) = {0}.
Proof. First note that given an algebraic vector bundle over a compact
complex manifold, any endomorphism of this bundle necessarily has
constant rank, since the coefficients of the characteristic polynomial
are regular functions and hence constant. Therefore we can talk of the
kernel and image of such an endomorphism as algebraic subbundles.
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS
7
Now, given a parabolic endomorphism f of a parabolic vector bundle
E, the subbundle ker(f ) has a canonical induced parabolic structure,
coming from the following short exact sequence
(2)
0 → ker(f ) → E → im(f ) → 0.
The same is true for im(f ), which we also think of as a subbundle
of E. One can easily see that the parabolic weights that ker(f ) and
im(f ) inherit as subbundles of E, when counted with multiplicities, are
complementary to each other with respect to the parabolic weights of
E.
Assume now that f is neither zero nor an isomorphism, so both
ker(f ) and im(f ) are proper subbundles of E. Using the stability we
have that
pdeg (ker(f ))
pdeg(E)
pdeg (im(f ))
pdeg(E)
(3)
<
and
<
.
rk (ker(f ))
rk (E)
rk (im(f ))
rk (E)
Using the complementarity of the parabolic weights of ker(f ) and
im(f ) however we also have that
(4)
pdeg(E)
pdeg (ker(f )) + pdeg (im(f ))
=
.
rk (E)
rk (ker(f )) + rk (im(f ))
One easily sees that the combination of (3) and (4) would give
pdeg (im(f ))
pdeg (im(f ))
pdeg (ker(f ))
pdeg (ker(f ))
<
and
<
,
rk (ker(f ))
rk (im(f ))
rk (im(f ))
rk (ker(f ))
hence f is either zero or an isomorphism.
If f is an isomorphism, one just has to take a point x ∈ X and
consider an eigenvalue λ of fx : Ex → Ex . Next consider the parabolic
endomorphism of E given by (f − λ IdE ), by the same reasoning as
before this is an isomorphism or it is zero. Hence H 0 (ParEnd(E)) = C.
In the case of a strongly parabolic endomorphism, as for any marked
point p ∈ D zero is one of the eigenvalues of fp , one gets similarly
H 0 (SParEnd(E) = 0.
Finally, notice that ParEndE is naturally dual to SParEnd(E)(D),
and vice versa SParEnd(E) is dual to ParEnd(E)(D). Throughout
the paper we shall often use Serre duality for the hypercohomology of a
complex on a curve, so we recall its statement: for a bounded complex
C of locally free sheaves on X of the form
0 → C0 → . . . → Cm → 0
we have the natural duality
Hi (C)∗ ∼
= H1−i+m (C ∗ ⊗ K),
8
MARINA LOGARES AND JOHAN MARTENS
where C ∗ ⊗ K is the complex
0 ← (C 0 )∗ ⊗ K ← . . . ← (C m )∗ ⊗ K ← 0.
2.2. Parabolic Higgs bundles. A parabolic Higgs bundle ([Sim90])
is a parabolic vector bundle together with a Higgs field Φ, a bundle
morphism
Φ : E → E ⊗ K(D),
where K is the canonical bundle of X, which preserves the parabolic
structure at each marked point:
Φ|p (Ep,i ) ⊂ Ep,i ⊗ K(D)|p ,
i.e. Φ ∈ H 0 (X, ParEnd(E) ⊗ K(D)). In keeping with the notation
introduced above we refer to the Higgs bundle as strongly parabolic if
the Higgs field is actually nilpotent with respect to the filtration, i.e.
if
Φ|p (Ep,i ) ⊂ Ep,i+1 ⊗ K(D)|p .
Similar to vector bundles a Higgs bundle is (semi) stable if the slope
condition
pdeg(F )
pdeg(E)
<
rk (F ) (=) rk (E)
holds, restricted now to all proper subbundles F preserved by the Higgs
field, i.e. with Φ(F ) ⊂ F ⊗ K(D).
Denote by Pα the moduli space of α stable parabolic Higgs bundles of
degree d and rank r, which was constructed by Yokogawa in [Yok93] and
further discussed in [BY96]. This space is a normal, quasi-projective
variety of dimension
(5)
dim Pα = (2g − 2 + n)r2 + 1.
Observe that this dimension is independent of the flag-type at the
marked points, in contrast to the dimension of Nα . Indeed, there is
a natural partial orderings on the flag types, and if Pα̃ is the corresponding moduli space for a finer flag type and the weights α̃and α
are close enough such that semi-stability is preserved by the forgetful
functor, then there is a forgetful morphism
(6)
Pα̃ → Pα
which is generically finite - see also proposition 2.2 below.
2.3. The Hitchin fibration.
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS
9
2.3.1. Hitchin map. Just as for ordinary Higgs bundles [Hit87a, Hit87b],
the parabolic Higgs bundles form an integrable system by means of the
Hitchin map, defined as follows. Given a vector bundle E, any invariant, homogeneous degree i polynomial naturally defines a map
H 0 (End(E) ⊗ K(D)) → H 0 (K(D)i ).
Now, take the elementary symmetric polynomials as a homogeneous basis of polynomials on gl(r) invariant under the adjoint action of GL(r),
then the corresponding maps ai combine to give the Hitchin map
hα : Pα → H
where the vector space H is the Hitchin space
H = H 0 (X, K(D)) ⊕ H 0 (X, K(D)2 ) ⊕ · · · ⊕ H 0 (X, K(D)r ).
The components of h are defined as follows: for any parabolic Higgs
bundle (E, Φ) and for any x ∈ X, let k ∈ K(D)|x . Then we have that
det(k.IdE|x − Φ|x ) = k r + a1 (Φ)(x)k r−1 + · · · + ar−1 (Φ)(x)k + ar (Φ)(x),
and hα (E, Φ) is given by (a1 (Φ), . . . , ar (Φ)). Notice that h is blind to
the parabolic structure at each marked point, as it only depends on Φ
and the line bundle K(D). In [Yok93, §5] it is shown that hα is proper,
and in fact projective.
2.3.2. Spectral curves. For each element s = (s1 , . . . , sr ) of H one can
define a spectral curve Xs in S, the total space of K(D), as follows:
pull back K(D) to S and denote its canonical section as λ. Then Xs
is the zero-locus of
λr + s1 λr−1 + · · · + sr ,
a (possibly ramified) covering of X. As usual, by a Bertini argument,
for a generic element in H the corresponding Xs is smooth. The genus
of Xs can be given using the adjunction formula:
2g(Xs ) − 2 = deg(KXs )
= KXs .Xs
= (KS + Xs ).Xs
= KS .Xs + Xs2
= rc1 (O(−D)) + r2 X 2
= −rn + r2 (2g − 2 + n)
and hence
(7)
g(Xs ) =
−rn + r2 (2g − 2 + n) + 2
.
2
10
MARINA LOGARES AND JOHAN MARTENS
The eigenvalues of Φ|x for x ∈ X control the ramification of Xs over
x, e.g. if all eigenvalues are 0, then Xs is completely ramified over x, if
all are different then Xs is unramified over x. We denote the covering
by ρ : Xs → X, with ramification divisor R on Xs .
2.3.3. Generic fibers. Now, if Xs is smooth, pull back E to Xs by ρ. We
canonically get a line bundle L on Xs , such that L(−R) sits inside this
pull back (see e.g. [BNR89, Prop.3.6]). Away from a ramification point
the fiber of L is given by an eigenspace of ρ∗ E, exactly corresponding
to the eigenvalue of Φ given by that point of Xs , in fact one has that
(8)
ρ∗ Φ−λId
0 → L(−R) → ρ∗ E −→ ρ∗ (E ⊗ K(D)) → L ⊗ ρ∗ K(D) → 0
is exact [BNR89, Remark 3.7]. Furthermore we have ρ∗ L = E, and the
Higgs field Φ is also easily recovered: multiplication by the canonical
section λ of ρ∗ K(D) descends to a morphism
Φ : ρ∗ L → ρ∗ L ⊗ K(D) = ρ∗ (L ⊗ ρ∗ K(D)) .
In order to obtain the degree of L, apply Grothendieck-RiemannRoch to the morphism ρ : Xs → X, and then integrate both sides.
This gives
1
1
deg(E) + r deg(TX ) = deg(L) + deg(TXs )
2
2
and hence
1
1
deg(L) = deg(E) + r deg(TX ) − deg(TXS )
2
2
2
= d + r(1 − g) + r (g − 1)
= d + r(1 − r)(1 − g).
Moreover, the smoothness of the spectral curve guarantees that only
finitely many parabolic structures are compatible with the Higgs field,
even if the Higgs field has repeated eigenvalues at the points of D.
Indeed, we have
Proposition 2.2. If the spectral curve Xs is smooth, then h−1
α (s) consists of disjoint copies of the Jacobian of Xs , one for every partition of
the eigenvalues along the multiplicities of the flagtype.
Proof. We want to show that with each line bundle L on Xs of degree
d + r(1 − r)(1 − g) we get a parabolic Higgs bundle of degree d.
As stated above, the push forward ρ∗ L determines a vector bundle
E on X [BNR89], and multiplication by λ on L descends to a Higgs
field. What remains is to construct a flag of the desired type at the
marked points. If p is in D, denote the eigenvalues of Φ|p by σ1 , . . . , σr .
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 11
Choose a partition of the σi according to the multiplicities mi (p) and
relabel the σi such that the partition is given by:
{σ1 , . . . , σm1 (p) }, {σm1 (p)+1 , . . . , σm1 (p)+m2 (p) }, . . . , {σr−mr(p) , . . . , σr }.
Now, choose a Zariski open set W around p such that L is trivial over
its inverse image ρ−1 (W ), and K(D) is trivial over W . Observe that
we can always do this: take any rational section of L, then by the
independence of valuation theorem (see e.g. [Deu73, page 19]) we can
choose a rational function such that the product of the two has no
poles or zeroes on ρ−1 (p). This new rational section trivializes L on
the complement of its divisor, which then easily gives the desired W .
Restricting to W , ρ∗ L is given as an O(W )-module by
O(W )[x]/(xr + s1 xr−1 + · · · + sr ),
with the si ∈ O(W ). Because of the choices made, we also know that
at p, the sj are given by the elementary symmetric polynomials in the
σi . Therefore the fiber of ρ∗ (L) over p is exactly given by
C[x]/((x − σ1 ) . . . (x − σr )).
Φ is of course given by multiplication
by x here, and hence this deterQ
(x
− σl ), with e1 = 1, then with
mines a basis: if we put ej = j−1
l=1
respect to this basis Φ looks like the lower-triangular matrix


σ1

 1 σ2


..


.
1
(9)
.



.
.
.. ..


1 σr
Hence by using e1 , . . . , er as an adapted basis, i.e.
Ep,1 =< e1 , . . . , er >, · · · , Ep,r(p) =< er−mr(p) (p) , . . . , er >,
we get the parabolic structure, which is independent of our trivializations of L and K(D) and the choice of the basis ei . From the matrix
form (9) one can immediately see that the eigenspace for every eigenvalue is one-dimensional, even if the eigenvalue has multiplicity, this is
what gives the uniqueness of the filtration.
It remains to show that this parabolic Higgs bundle is stable: for
this observe that a Φ-preserved subbundle of E would necessarily correspond to a subsheaf of L (see again [BNR89, page 174]), and as Xs is
assumed to be smooth, this has to be a locally free sheaf itself as well,
necessarily of lower degree. Notice that this shows that smoothness of
the spectral curve implies that there are no subbundles preserved by
12
MARINA LOGARES AND JOHAN MARTENS
the Higgs-field, and hence the slope-stability condition need not even
be applied.
In section 3.2.4 we shall see that each of these Jacobians is actually
contained in a different symplectic leaf for the Poisson structure on Pα .
3. Poisson structure
3.1. Background material. We will begin by briefly reviewing the
background on Poisson geometry and Lie algebroids that we need. Remark that in the literature Lie algebroids and groupoids are usually
used in a differential geometric setting. For our purposes, where we
will use these notions on the moduli space of parabolic Higgs bundles,
all structures (bundles, spaces, actions) are algebraic however.
There are two main differences in the holomorphic or algebraic setting vs. the smooth settings: algebroids have to be defined using the
sheaf of sections of the underlying vector bundle rather than just the
global sections, and more significantly, principal bundles do not always
have connections, or equivalently the corresponding Atiyah sequence
(15) does not always split. As we only use Lie groupoids and algebroids
in a smooth setting, we have however kept the differential geometric
notions of submersion etc. For more background on Lie groupoids and
algebroids see [CdSW99, Mac05], which we use without reference in
this section.
3.1.1. Poisson geometry. There are many ways of packaging a Poisson structure, so just to fix conventions we shall state the one most
convenient for our purposes:
Definition 3.1. A Poisson structure on a complex manifold M is given
by a bundle morphism ] : T ∗ M → T M that is anti-symmetric, i.e. ]∗ =
−], such that the Schouten-Nijenhuis bracket [˜], ˜]] of the corresponding
V
bivector ˜] ∈ 2 T M is zero.
Now, for any manifold N , look at the total space of the cotangent
bundle π : T ∗ N → N . The tangent and co-tangent bundles to T ∗ N
both fit in short exact sequences,
(10)
0 → π ∗ (T ∗ N ) → T ∗ (T ∗ N ) → π ∗ T N → 0
and
(11)
0 → π ∗ (T ∗ N ) → T (T ∗ N ) → π ∗ T N → 0.
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 13
The canonical Poisson structure (which is of course even a symplectic
structure) is defined as the unique anti-symmetric bundle morphism
]T ∗ N : T ∗ (T ∗ N ) → T (T ∗ N ) such that in the diagram
(12)
0
/
Id
0
/
π ∗ (T ∗ N )
/
/
π ∗ (T ∗ N )
/
T ∗ (T ∗ N )
/
0
]T ∗ N
/
0
π∗T N
−Id
/
T (T ∗ N )
π∗T N
all squares commute.
Definition 3.2. A morphism f : M1 → M2 between Poisson spaces
(M1 , ]1 ) and (M2 , ]2 ) is Poisson if the square
(13)
T ∗M
O 1
]1
/
(df )∗
T M1
T ∗ M2
/
]2
df
T M2
commutes.
In particular, if a Poisson manifold M is equipped with an action
by a group G that preserves the Poisson structure, then if the quotient
exists the quotient map M → M/G is a Poisson morphism.
3.1.2. Algebroids and Poisson structures. Two places where Poisson
structures naturally occur are on the quotients of a symplectic manifold by a Hamiltonian group action - where the symplectic leaves are
given by the various symplectic reductions. Another place is on the
total space of the dual of a Lie algebroid (which includes as a special
example the dual of a Lie algebra). We will mainly be interested in
the special case of an Atiyah algebroid, which is an example of both of
these situations.
Definition 3.3. A Lie algebroid over a (complex) variety M is a vector
bundle E → M such that the sheaf of sections of E is a sheaf of Liealgebras for a bracket
[., .] : O(E)(U ) × O(E)(U ) → O(E)(U ),
together with a bundle map, the anchor, a : E → T M which preserves
the Lie-brackets on sections. Moreover the following Leibniz rule has
to hold, for f ∈ O(U ), X, Y ∈ O(E)(U ) :
[X, f Y ] = f [X, Y ] + (a(X)f )Y.
14
MARINA LOGARES AND JOHAN MARTENS
Two natural classes of examples of Lie algebroids are given by tangent bundles T M of a manifold M (where the map a is the identity),
and Lie algebras g, regarded as a vector bundle over a point. One can
think of transitive Lie algebroids (i.e. algebroids with surjective anchor
maps) as interpolations between these two.
The relevance of Lie algebroids for us is through the following theorem, which in this form is due to Courant [Cou90] (see also [Mac05,
CdSW99] or [Bot95]):
Theorem 3.4. The total space of the dual vectorbundle E ∗ of a Lie
algebroid E has a natural Poisson structure.
For the two previous examples mentioned above the Poisson structures are given by the canonical symplectic structure on the total space
of T ∗ M , and the Kirillov-Kostant-Souriau Poisson structure on g∗ . In
the former there is one single symplectic leaf, in the latter case the
symplectic leaves are given by the co-adjoint orbits of the Lie group G.
Let a group G act freely and properly on a manifold P , in other words
π
P → P/G is a G-principal bundle. Of course then G also acts freely and
properly, in a Hamiltonian fashion, on the symplectic manifold T ∗ P ,
and therefore the quotient T ∗ P/G is a Poisson manifold. Another way
to realize the Poisson structure on T ∗ P/G is as the dual of a particular
type of Lie algebroid, the so-called Atiyah algebroid, as follows. G acts
freely on T P , and T P/G is a Lie algebroid over P/G. One sees this
most easily by interpreting the sections of T P/G as G-invariant vector
fields on P , and the sections of T (P/G) as G-invariant sections of the
bundle on P that is the quotient of T P by the tangent spaces to the
orbits:
(14)
0 → Torbits P → T P → π ∗ T (P/G) → 0.
The anchor map a : T P/G → T (P/G) is just given by projecting an
invariant vectorfield to the part ‘orthogonal’ to the orbits. It is clear
that this satisfies the required property, since functions on P/G correspond to G-invariant functions on P , and any tangent field along the
orbits annihilates an invariant function.
All in all the above shows that T P/G is the extension of T (P/G)
by the adjoint bundle Ad(P ) = P ×Ad g. The latter is, as a bundle of
Lie algebras of course a Lie algebroid with trivial anchor map, and the
corresponding Atiyah sequence
(15)
0 → Ad(P ) → T P/G → T (P/G) → 0
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 15
preserves all Lie brackets on local sections.
Though mainly used in a differential geometric setting, the Atiyah
algebroid and the corresponding short exact sequence was originally
introduced in [Ati57] in the context of the study of the existence of
holomorphic connections in complex fiber bundles.
3.1.3. Symplectic leaves for the dual of an algebroid. Since we are concerned with Atiyah algebroids we can study the symplectic leaves on
its dual fairly directly, using the general fact that the symplectic leaves
of the Poisson reduction of a symplectic manifold by a free Hamiltonian group action correspond to the various symplectic reductions. We
shall do this in section 3.2.4. For the sake of completeness we do briefly
indicate here however that these symplectic leaves can be seen as coadjoint orbits of a Lie groupoid. For us this is mainly of philosophical
relevance, leading to the interpretation of Pα in the case of full flags as
a Grothendieck-Springer resolution.
Just as Lie algebras g are given as tangentspaces to Lie groups G, Lie
algebroids can come from a Lie groupoid - though not every algebroid
integrates to a Lie groupoid, see [CF03].
Definition 3.5. A Lie groupoid G ⇒ M over a manifold M is a space3
G together with two submersions α, β : G → M , as well as an associative product (g1 , g2 ) 7→ g1 g2 defined on composable pairs, i.e. (g1 , g2 )
s.t. β(g2 ) = α(g1 ), such that α(g1 g2 ) = α(g2 ) and β(g1 g2 ) = β(g1 ).
Furhermore an identity section : M → G has to be given, such that
the following hold for all g ∈ G:
(β(g)) = g and (α(g)) = g
as well as an inversion ι : G → G, with
ι(g)g = (β(g)) and gι(g) = (α(g)).
The maps α and β are often referred to as respectively the source
and target maps of the groupoid, and one thinks of G as consisting of
arrows g from α(g) to β(g), and compositions, inverses and identities
can be understood as such.
With every Lie groupoid one can naturally associate a Lie algebroid structure on the normal bundle to M ∼
= (M ) ⊂ G. We refer
3In
the differential geometric setting it is in general not required here that G is
Hausdorff, but it is assumed that M is.
16
MARINA LOGARES AND JOHAN MARTENS
to [CdSW99, Mac05] for further background material.
Now, let G ⇒ M be a Lie groupoid, with associated Lie algebroid
E → M . Then there exists a cotangent groupoid T ∗ G ⇒ E ∗ , which
is both a vector bundle over G and a Lie groupoid over E ∗ . Clearly
any groupoid H ⇒ N acts on its base N . Of particular relevance for
us is the following (see e.g. [Mac05, Proposition 11.5.4 and Theorem
11.5.18]):
Theorem 3.6. The symplectic leaves for the Poisson structure on the
total space of the dual of a Lie algebroid E → M associated with a Lie
groupoid G ⇒ M are the (connected components of ) the orbits for the
action of T ∗ G ⇒ E ∗ on E ∗ .
The orbits of T ∗ G ⇒ E ∗ are often referred to as the co-adjoint orbits
of the original groupoid G ⇒ M .
This theorem establishes a cotangent groupoid as a particular case of
the general notion of a symplectic groupoid (see [CDW87]). The base
of a symplectic groupoid is always Poisson, and its symplectic leaves
are given by the orbits of the symplectic groupoid [CDW87].
3.2. Poisson structure on Pα .
3.2.1. The ] map. The tangent space to Pα at a stable parabolic Higgs
bundle (E, Φ) is given by the H1 hypercohomology of the two-term
complex
(16)
[. ,Φ]
ParEnd(E) −→ ParEnd(E) ⊗ K(D).
Let us now write down the Poisson bracket. The dual of the complex
(16), tensored with K, is given by
(17)
−[. ,Φ]
SParEnd(E) −→ SParEnd(E) ⊗ K(D).
We can now inject (17) into (16), as follows:
(18)
SParEndE
−[. ,Φ]

/
Id
−Id⊗Id

K(D)
SParEnd(E) ⊗ K(D) /
ParEnd(E)
[. ,Φ]
ParEnd(E) ⊗ K(D).
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 17
Using Serre duality for hypercohomology we therefore get a map
∗
(19) ]Pα : T[E,Φ]
Pα ∼
= H1 (SParEnd(E) → SParEnd(E) ⊗ K(D))
→ H1 (ParEnd(E) → ParEnd(E) ⊗ K(D)) ∼
= T[E,Φ] Pα .
Because of the choice of signs in (18) ]Pα is antisymmetric. We now
want to show that this determines a Poisson-structure on Pα . One
could try to do this directly by calculating the Schouten-Nijenhuis
bracket, but it would be rather hard and not so instructive, therefore
we follow a different route below.
3.2.2. Poisson structure via Lie algebroids. Now let Pα0 be the open
subvariety of Pα consisting of those parabolic Higgs bundles (E, Φ)
whose underlying parabolic bundle is stable. As mentioned before,
this is a vector-bundle over Nα , the moduli space of parabolic vector
bundles, with fiber H 0 (X, ParEnd(E) ⊗ K(D)).
We shall use the following two projections:
(20)
Nα
Nα × X
H
uu
uu
u
uu
uz u
η
HH
HHν
HH
HH
#
X
Now, on Nα × X we have a parabolic universal bundle [BY99, Theorem 3.2], which we denote by E. This leads to a short exact sequence
of sheaves on Nα × X:
Y
lp ⊗ Oν −1 (p) → 0.
0 → SParEnd(E) → ParEnd(E) →
p∈D
Applying η∗ to this sequence gives the exact sequence
Y
(21) 0 → η∗ SParEnd(E) → η∗ ParEnd(E) → η∗
lp ⊗ Oν −1 (p) →
Y
R1 η∗ SParEnd(E) → R1 η∗ ParEnd(E) → R1 η∗
lp ⊗ Oν −1 (p) → 0.
Q
As the support of lp ⊗Oν −1 (p) has relative dimension zero with respect
to η, the last term of this sequence is easily seen to be zero by relative
dimension vanishing (see e.g. [Har77, III.11.2]). The first term is zero
since every stable bundle is simple (lemma 2.1), and for the same reason
the second term4 is an invertible sheaf. Moreover, R1 η∗ ParEnd(E) is
the tangent sheaf to Nα , hence we shall denote it as TNα .
4Observe
that this term would also be zero in the case of a semi-simple structure
group, so for parabolic principal Higgs bundles the relevant principal bundle over
the moduli space Nα would just have the product of the Levi groups as structure
group, without quotienting by global endomorphisms.
18
MARINA LOGARES AND JOHAN MARTENS
Lemma 3.7. All of the sheaves occuring in the sequence (21) are locally
free.
Proof. It suffices to notice that the corresponding cohomology groups
(e.g. H 1 (SParEnd(E)) have constant rank as E varies in Nα , and
apply Grauert’s theorem [Har77, III.12.9].
Q
Denote now η∗
lp ⊗ Oν −1 (p) /η∗ ParEnd(E) by Ad. This clearly
is a bundle of Lie algebras. Our claim is that the short exact sequence
(22)
0 → Ad → R1 η∗ SParEnd(E) → TNα → 0
is an Atiyah sequence, as in (15).
Lemma 3.8. The dual of the vector bundle Pα0 → Nα is R1 η∗ SParEnd(E).
Proof. By relative Serre duality for the morphism η in (20), the dual
of R1 η∗ SParEnd(E) is given by η∗ ParEnd(E ⊗ q ∗ K(D)), which is a
locally free sheaf as well, with fibers of the associated vector bundle
over a point [E] ∈ Nα given by H 0 (ParEnd(E ⊗ K(D))). Clearly the
obvious map bundle morphism η∗ (SParEnd(E ⊗ q ∗ K(D)) → Pα0 is an
isomorphism.
We now assume the existence of a principal bundle π : Fα → Nα
with structure group
!
Y
(23)
L=
Lp /C,
p∈D
where C is the diagonal subgroup of the product of the centers of the
Pp . The total space of Fα can be interpreted as a moduli space of
α-stable framed parabolic bundles, where a framed parabolic bundle
is a parabolic vector bundle together with a framing of the associated
graded space at the marked points, i.e. the choice of an isomorphism
Gr(p) =
r(p)
M
∼
=
Fp,i /Fp,i+1 −→
M
Cmi (p)
i=1
at each marked point p ∈ D. In appendix A a construction for Fα is
given in the case of full flags.
Theorem 3.9. The sequence (22) is the Atiyah sequence for the L
principal bundle Fα → Nα .
Proof. Since the sheaf of sections of the adjoint bundle of a principal
bundle is the direct image of the relative tangent sheaf of the associated
projection, it suffices to show that the inverse image under π of the
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 19
short exact sequence (14) is the sequence on Fα determining the relative
tangent sheaf:
0 → Tπ → TFα → π ∗ TNα → 0.
To fix notation, let us look at the commutative diagram
ηe
/
π
e
Fα × X
(24)
Nα × X
η
/
π
Fα
Nα
where Fα is a moduli space of framed stable parabolic bundles. The
space Fα × X comes equipped with a sheaf of framed endomorphisms
of a universal bundle FParEnd, such that the tangent sheaf to Fα is
given by R1 ηe∗ FParEnd. As it is also easy to see that FParEnd is
equal to π
e∗ SParEnd(E), the commutativity of the diagram (24) - and
flatness of π, guaranteed by invoking e.g. [EGA, proposition 6.1.5] give that indeed
R1 ηe∗ FParEnd ∼
= π ∗ R1 η∗ SParEnd((E)),
see e.g. [Har77, Proposition III.9.3].
3.2.3. Extension of the bracket.
Theorem 3.10. The bivector on Pα determined by (19) extends the
Poisson structure on Pα0 given by the Lie algebroid on (Pα0 )∗ .
Proof. For an α-stable framed parabolic bundle (E, ∼
=) the tangent and
cotangent spaces to Fα are given respectively by
1
T[E,∼
=] Fα = H (SParEnd(E))
and
∗
0
T[E,
∼
=] Fα = H (ParEnd(E) ⊗ K(D)).
The tangent and the cotangent spaces to T ∗ Fα at a point [E, Φ, ∼
=] are
given by the first hypercohomology groups
,Φ]
∗
∼ 1 SParEnd(E)[.−→ParEndE⊗K(D)
T[E,Φ,∼
=] T Fα = H
and
−[. ,Φ]
∗
∗
1
∼
T[E,Φ,
T
F
H
SParEnd(E)
−→
ParEndE⊗K(D)
.
=
∼
α
=]
Furthermore, if we look at the short exact sequences of complexes
/
0
0
/
/
0
ParEnd(E) ⊗ K(D)
SParEnd(E)
/
±[, ,Φ]
ParEnd(E) ⊗ K(D)
/
/
0,
/
0
SParEnd(E)
/
0
20
MARINA LOGARES AND JOHAN MARTENS
take its long exact sequence in hypercohomology, and notice that
H 0 (SParEnd(E)) is zero since E is stable as a parabolic vector bundle,
we obtain the short exact sequences
SParEnd(E)
0
1
±[. , Φ] ↓
0 → H (ParEnd(E) ⊗ K(D)) → H
ParEnd(E) ⊗ K(D)
→ H 1 (SParEnd(E)) → 0,
corresponding to (10) and (11). Now, using the characterization given
in (12) for the canonical Poisson structure on T ∗ Fα , one can see that
this is induced by the morphism of complexes
SParEndE
−[. ,Φ]

/
Id
SParEnd(E)
[. ,Φ]
−Id⊗Id

K(D)
ParEnd(E) ⊗ K(D) /
.
ParEnd(E) ⊗ K(D).
Since the map T ∗ Fα → Pα0 is a Poisson morphism, and using the
characterization (13) and the definition (19) of ]Pα it suffices to notice
that by these choices indeed the square
SParEnd(E)
SParEnd(E)
1
1
/H
−[. , Φ] ↓
[. , Φ] ↓
H
ParEnd(E) ⊗ K(D)
ParEnd(E) ⊗ K(D)
O
H
1
SParEnd(E)
−[. , Φ] ↓
SParEnd(E) ⊗ K(D)
/
H
1
ParEnd(E)
[. , Φ] ↓
ParEnd(E) ⊗ K(D)
commutes.
As Pα0 is open and dense in Pα this establishes the Poisson structure
on all of Pα .
3.2.4. Symplectic leaves. Given a principal G-bundle P → M , the symplectic leaves of the dual of an Atiyah algebroid T P/G are simply the
symplectic reductions of the cotangent bundle T ∗ P . The lift of the action of a group G on a manifold N to the total space of the cotangent
bundle T ∗ N is of course always Hamiltonian, with a canonical moment
map given by
g ∈ g, χ ∈ Tx∗ N :
µ(χ)(g) = χ(ξg (x)),
where ξg is the Hamiltonian vector field corresponding to g ∈ g. In the
particular case where the action of G is free, i.e. when the manifold
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 21
is a principal G-bundle, the moment map can also be understood by
dualizing the sequence (14):
µ
∗
0 → π ∗ T ∗ (P/G) → T ∗ P → Torbits
→ 0,
∗
P ∼
observing that Torbits P ∼
= P × g∗ . Recalling (21)
= P × g and Torbits
and the proof of Theorem 3.9 this tells us immediately that on T ∗ Fα
the moment map µ is given by the parabolic residue, i.e. the map that,
when restricted to a fiber over a framed parabolic bundle E, gives
!
M
H 0 (SParEnd(E) ⊗ K(D)) → ker
l∗p → H 0 (ParEnd(E))∗
p∈D
which is the dual of the boundary map associated with the short exact
sequence
M
0 → SParEnd(E) → ParEnd(E) →
lp ⊗ Op → 0.
p∈D
The reduction, and hence symplectic leaves, are simply µ−1 (O)/L,
where O is a co-adjoint orbit in Lie(L). Notice that in the generic
case, when the eigenvalues of the Higgs field at the marked points are
all distinct, there is a unique co-adjoint orbit with these eigenvalues.
When eigenvalues are repeated on a particular Ep,i /Ep,i+1 there will be
several co-adjoint orbits.
One can check that this agrees with the rank of the Poisson structure
given by ]Pα at a parabolic Higgs bundle (E, Φ). Indeed, if one recalls
the definition (19) then the short exact sequence of complexes
/
0
SParEnd(E)
−[. ,Φ]
0
/
SParEnd(E) ⊗ K(D)
/
Id
[. ,Φ]
/
/
ParEnd(E)
ParEnd(E) ⊗ K(D)
/
L
L
lp ⊗ K(D)p
ParEnd(E) ⊗ K(D)
and hence the generic, maximal rank of ]Pα , occuring when all eigenvalues of Φ are different, is
(25)
rk Pα = dim Pα − nr + 1 = (2g − 2)r2 + nr(r − 1) + 2.
0
/
0
[. ,Φ]|Gr(p)
gives rise to the long exact sequence of hypercohomology (at least when
E is stable):
SParEnd(E)
ParEnd(E)
]Pα
1
1
↓
↓
0 → C → ⊕ ker [. , Φ]|Gr(p) → H
−→ H
SParEnd(E) ⊗ K(D)
/
lp ⊗ O p
→ ...,
22
MARINA LOGARES AND JOHAN MARTENS
The eigenvalues of the Higgs field are in fact determined by the
Hitchin map, and as the latter is blind to the parabolic structure, it
factors through morphisms of the form (6):
Pα̃ @
(26)
@@ h
@@ α̃
@@
@
Pα h / H
α
e
/
H/H0 ,
where H0 is the subspace of H given by
H0 = H 0 (X, K) ⊕ H 0 (X, K 2 (D)) ⊕ · · · ⊕ H 0 (X, K r (Dr−1 )).
Observe that the Hitchin map for the moduli space of strongly parabolic
Higgs bundles (which is a symplectic leaf of Pα ) takes values in H0 . One
easily sees that, roughly speaking, H/H0 determines the eigenvalues of
the Higgs field at the marked points p ∈ D, without ordering. In the
case where α̃ corresponds to full flags, the connected components (one
for every ordering of the eigenvalues) of the fibers of e ◦ hα̃ are exactly
the symplectic leaves.
3.2.5. Complete integrability of the Hitchin system. The composition
e ◦ hα from (26) also plays a role in the Hitchin system, which we can
discuss now that we have the Poisson structure at our disposal. Recall
that for a holomorphic Poisson manifold or Poisson variety of dimension 2k + l, where the rank (or dimension of the generic leaf) of the
Poisson structure is 2k, a completely integrable system is given by k + l
Poisson-commuting, functionally independent functions, such that l of
them are Casimirs, i.e. they Poisson-commute with any function. Furthermore the generic fiber is required to be an Abelian variety. The
connected components of the simultaneous fibers of the Casimir functions are the closures of the top-dimensional symplectic leaves.
By Riemann-Roch one gets (assuming that n ≥ 1)
dim(H) = (2g − 2 + n)
(r + 1)r
+ r(1 − g)
2
and
dim(H0 ) = (2g − 2)
r(r + 1)
r(r − 1)
+n
+ r(1 − g) + 1,
2
2
hence
dim (H/H0 ) = nr − 1.
Notice also from (7) that g(Xs ) = dim (H0 ), and from (5) that
2 dim (H0 ) + dim (H/H0 ) = dim Pα .
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 23
In order to show directly that Pα equipped with hα is a completely
integrable system, we would have to work with local information, in
order to establish the vanishing of the relevenant Poisson brackets.
As our map ]Pα is defined fibre-wise however, we use an alternative
characterization (following [Mar94, Section 8.1]):
Proposition 3.11. The generic fiber h−1
α (s), corresponding to a smooth
spectral curve Xs , is Lagrangian in a symplectic leaf for the Poisson
structure on Pα .
In order to prove this we shall need two lemmas. The first is the
following characterization of coisotropic submanifolds of symplectic
leaves, which is easy to see:
Lemma 3.12. Let J be a submanifold of a symplectic leaf L of a Poisson manifold (M, ]). Then J is co-isotropic in L if for any point p ∈ J
we have that
](Np∗ J) ⊂ Tp J,
where Tp J is the tangent space to J at p, Np J is the normal space at p
of J in M , and Np∗ J is the conormal space at p.
The second is a description of the tangent space to h−1
α (s):
Lemma 3.13. Let (E, Φ) be a Higgs-bundle in one of the components
of h−1
α (s), which we identify with the Jacobian Js of Xs . Then we have
the following short exact sequence on X:
ParEnd(E)
1
1
∼
[. , Φ] ↓
0 → T(E,Φ) Js = H (ρ∗ OXs ) → H
→ H 0 ((ρ∗ KXs )(D)) → 0
ParEnd(E) ⊗ K(D)
Proof. If we tensor the exact sequence (8) on Xs with L−1 (R), push it
forward by ρ and use Hurwitz’ theorem, we obtain the exact sequence
[. ,Φ]
0 → ρ∗ (OXs ) → ParEnd(E) −→ ParEnd(E)⊗K(D) → ρ∗ (KXs )(D) → 0
on X. Using this we can look at the short exact sequences of complexes
0
/ ρ∗ (OX )
s
/
[. ,Φ]
/
0
/
0
/
ParEndE
ParEnd(E) ⊗ K(D)
/
ParEnd(E) ⊗ K(D)
and
0
/
im([. , Φ])
0
/
im([. , Φ])
/
/
im([. , Φ])
ParEnd(E) ⊗ K(D)
/
0
/
0
im([. , Φ])
/
0
/ ρ∗ (KX )(D)
s
/
/
0
0
24
MARINA LOGARES AND JOHAN MARTENS
Combining the hypercohomology long exact sequences gives the desired
result.
Proof of Proposition 3.11. By the discussion in section 3.2.4, it is clear
that h−1
α (s) is contained in a symplectic leaf. Notice that by (7), Proposition 2.2 and (25) we already know that it suffices to show that h−1
α (s)
is co-isotropic. With this we can now apply Lemma 3.12 to a connected
component of h−1
α (s) corresponding to a smooth spectral curve Xs by
observing that in the commutative diagram
/
H 0 (ρ∗ KXs (D))∗ ∼
= H 1 (ρ∗ OXs (−D))
1
H
−[. ,Φ]
SParEnd(E) → SParEnd(E)⊗K(D)
H 1 (ρ∗ OXs )∗ ∼
= H 0 (ρ∗ KXs )
]Pα
/
H 1 (ρ∗ OXs )
1
H
[. ,Φ]
/
H 0 (ρ∗ KXs (D))
the columns (given by Lemma 3.13) are exact. This ends the proof of
Proposition 3.11.
This finally gives us:
Theorem 3.14. The moduli spaces Pα , with the Poisson structure
introduced above and the Hitchin map hα , forms a completely integrable
system, for which the Casimirs are given by e ◦ hα .
4. Morphisms between moduli spaces,
Grothendieck-Springer resolution
Given a complex semi-simple connected Lie Group G with Lie algebra g and Weyl group WG , one can construct the so-called GrothendieckSpringer morphism. There are various incarnations of this, for the
group, the Lie algebra, etc, so we just briefly recall this here. The
Grothendieck-Springer space is defined as
GS G = {(g, b)|g ∈ g, b ∈ G/B, g ∈ b},
where B is a Borel subgroup5 of G. The obvious map µ : GS G → g is
widely used in geometric representation theory, see e.g. [CG97]. It is
generically finite (|WG | : 1), and provides a resolution of singularities of
the nilpotent cone n ⊂ g, which is referred to as Springer’s resolution.
After choosing an equivariant identification g ∼
= g∗ we can think of GS G
5One
Borels.
ParEnd(E) → ParEnd(E)⊗K(D)
can generalize this to G/P , that is, to parabolic subgroups other than
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 25
as the dual of an algebroid over G/B, and the map µ as the momentmap for the induced G action. In particular GS G is a regular Poisson
manifold (i.e. all the symplectic leaves have the same dimension), and
µ will be a Poisson morphism. Moreover there is the following diagram,
called the Grothendieck simultaneous resolution:
(27)
/
GS G
g
/
t
t/WG
where t is the abstract Cartan and WG the abstract Weyl group of G.
For more details regarding this we refer to [CG97, Section 3.1].
Our construction gives a similar picture for Atiyah algebroids rather
than Lie algebras, where the role of g is now played by any of the Pα ,
but in particular can be the moduli space of parabolic Higgs bundles
with minimal flag-type (see also section 5 below), and the role of the
Grothendieck-Springer variety by the moduli space of parabolic Higgs
bundles with full flags. Indeed, we show below easily that this and
similar forgetful morphisms are Poisson.
Let us look at the moduli spaces for two different flag types on the
same divisor of marked points, Pα̃ and Pα , where the flag type of the
latter is coarser than that of the former. We assume that the parabolic
weights α and α̃ are close enough that if one forgets part of the flag on
an α̃-stable parabolic Higgs bundle the result is α-stable, so that we
obtain a morphism
(28)
Pα̃ → Pα .
Proposition 4.1. The morphism (28) is Poisson.
Proof. Let us denote a parabolic Higgs bundle for the finer flag type
e Φ)and
e
as (E,
its image under the forgetful morphism as (E, Φ). Then
clearly we have the natural inclusions of sheaves
e
SParEnd(E) ⊂ SParEnd(E)
and
e ⊂ ParEnd(E).
ParEnd(E)
26
MARINA LOGARES AND JOHAN MARTENS
Therefore we get that the diagram
e
SParEnd(E)
1
↓
H
/
H
e ⊗ K(D)
SParEnd(E)
1
e
ParEnd(E)
↓
e ⊗ K(D)
ParEnd(E)
ParEnd(E)
↓
ParEnd(E) ⊗ K(D)
O
H
1
SParEnd(E)
↓
SParEnd(E) ⊗ K(D)
/
H
1
commutes.
As said above, it is particularly interesting to look at the morphism
(28) in the case where Pα̃ corresponds to full flags, as then Pα̃ is a
regular Poisson manifold. We can put things together in the analogue
of the Grothendieck simultaneous resolution (27):
(29)
?? h
?? α̃
??
?
hα /H
Pα
/
o
Pα̃ ?
/
H/H0
∼
=
/
rn
Crn /C
(C /(Sr )n ) /C.
The map o : Pα̃ → Crn → Crn /C is just given by the eigenvalues of
Φ at the marked points - because of the full flags they come with an
ordering.
5. Comparison with Bottacin-Markman and further
remarks
A particular case, the case of minimal flags, of the above has already
been discussed in the literature, in independent work by Bottacin [Bot95]
and Markman [Mar94], though it is not framed in terms of parabolic
(Higgs) bundles or algebroids. Reviews of this work also appeared in
[DM96, Don03]. Bottacin and Markman study stable pairs or twisted
Higgs bundles, i.e. a vector bundle E over a curve X together with
a morphism Φ : E → E ⊗ F , for some fixed line bundle F . Pairs of
this kind (working over a field of positive characteristic), their moduli
stack and the Hitchin fibration for them also played a crucial role in the
recent work of Laumon-Ngô [LN04]. A moduli space for these was constructed by Nitsure in [Nit91], and in [Bot95] and [Mar94] it is shown
that, if deg(F ) > deg(K) (or F = K) and once one chooses an effective
divisor D in F K −1 , this space has a canonical Poisson structure.
Once this choice is made, and if D is moreover reduced, such a staΦ
ble pair (E, E → E ⊗ K(D)) can of course also be interpreted as a
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 27
parabolic Higgs bundle for the minimal flag type E|p,1 ⊃ E|p,2 = {0}.
In the case of such minimal flags there is only a single weight at each
marked point, and one sees that it does not contribute to the slope inequality (1). Therefore one cannot afford the luxury of the assumption
of genericity of the weights, and unless the rank and degree are coprime
there are properly semi-stable points, and the moduli space of vector
bundles is singular. Even over the non-singular locus, corresponding to
the stable vector bundles, there does not exist a universal bundle, but
there is however still a sheaf over the stable locus playing the role of
sheaf of endomorphisms of a universal bundle [Bot95, Remark 1.2.3],
which is as useable as our ParEnd(E)6.
Both Bottacin and Markman are primarily focused on the Poisson structure, and make no mention of Lie algebroids. Nevertheless,
Bottacin uses the same philosophy of obtaining the Poisson structure
through studying the dual vector bundle. He even writes down the
definition of a Lie algebroid and proves theorem 3.4 in [Bot95, Section
4.2]. He however does not identify the algebroid as an Atiyah algebroid,
but rather exhibits the Lie bracket on local sections explicitly on the
level of cocycles and cochains.
Markman does use the principal bundle over the moduli space of
bundles (using a construction of Seshadri [Ses82]), but phrases everything in terms of reduction of its cotangent bundle. Despite this our
approach is closest to Markman, and a careful reader might find several
similarities in our exposition, in particular in section 3.2.5, for which
we were helped by [Mar94, section 8.1].
Neither Bottacin or Markman make the restriction that we do that
D is a reduced divisor - i.e. they allow the Higgs field Φ to have poles of
arbitrary order, when interpreted as a meromorphic bundle morphism
from E to E ⊗ K, though Bottacin assumes D to be reduced in some of
his proofs. Also for non-minimal flags this would be a desirable property, in particular in the light of the geometric Langlands program with
wild ramification, and should not be significantly more complicated.
Another obvious generalization would be to look at other semi-simple
or reductive structure groups, replacing the use of spectral curves with
6Notice
that in the case of parabolic bundles with non-generic weight one could
use the same strategy, in fact for some of the non-generic weights an actual universal
bundle does exist over the stable locus, see [BY99, Theorem3.2].
28
MARINA LOGARES AND JOHAN MARTENS
cameral covers. Though most of the statements we make can at least
formally be translated into this setting, we have refrained from working
in this generality as it seems that the dust has not settled on the
notion of stability for parabolic principal bundles, cf. [BR89, TW03,
BBN01, BBN03]. By working in the context of stacks rather than
moduli schemes these problems would of course be avoided, and we
intend to take up this matter in the future.
Appendix A. A Levi principal bundle over the moduli
space of parabolic bundles
In this appendix we give a construction of a principal bundle Fα →
Nα with structure group L, in the particular case of full flags at all
marked points. In the general case, one should think of the total space
Fα of this principal bundle as a moduli space for α-stable parabolic
bundles, together with isomorphisms of all consecutive quotients in the
flags to a fixed vector space
Ei (p)/Ei+1 (p) ∼
= Cmi .
One can think of several approaches to this problem.
One approach (for general flag types) one could take is suggested in
[Hu06]: start from a suitable moduli space of framed vector bundles
(also known as bundles with level structure, in the case where the divisor over which one frames is reduced), as was for instance constructed
in [HL95], generalizing earlier work by Seshadri [Ses82]. The structure
group of the vector bundles under consideration acts on this space by
changing the framing, and we would like to take a GIT-style quotient
by the unipotent radical of the parabolic subgroup. For the Borel subgroup (leading to full flags) this is described in [Hu06], a more general
approach is given in [DK07, Kir08a]. The various Nα would then be
given by GIT quotients by the Levi group of this space, with the α occuring as the choice of a linearization. As such the Nα are, for generic
α, geometric quotients for the Levi group actions, and therefore (with
some mild extra conditions) principal bundles for the Levi group, see
below. Notice that the action of P GL(r) on a moduli space of framed
bundles was discussed in [BGM08], where it was shown that the action
linearizes and the GIT quotient is the moduli space of vector bundles.
In order to keep the exposition from becoming too technical we shall
use a contruction already done in the literature, following [HJS05].
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 29
Here a projective variety F was constructed directly, which was interpreted as a moduli space of framed parabolic sheaves7. The construction was inspired by a similar construction [HJS06] in symplectic
geometry through symplectic implosion. The connection between symplectic implosion and non-reductive GIT was discussed in [Kir08b].
This variety F comes with a natural torus action, an action which linearizes on a relatively ample line bundle. Using an earlier construction
for the moduli space of parabolic bundles given by Bhosle [Bho89], it is
shown that at a linearization given by a character α, the GIT quotient
is the moduli space of parabolic vector bundles, F//α T ∼
= Nα (even
in the case of partial flags, if one uses the αi with the corresponding
multiplicities). If α is regular, i.e. we are looking at full flags, this is
sufficient for us: the Luna slice theorem [Lun73], see also [GIT, Appendix to Chapter 1] and [HL97, Corollary 4.2.13], now establishes that
the α-stable locus (the stability simply corresponds to the stability of
the underlying bundle) Fα ⊂ F is a principal bundle which is locally
trivial in the étale topology. Furthermore, by a result of Serre [Ser58],
as the structure group is a torus, it is even locally trivial in the Zariski
topology.
References
[Ati57]
[BBN01]
[BBN03]
[BGM08]
[Bho89]
[Biq91]
[BNR89]
[Bot95]
[BR89]
7In
M. F. Atiyah. Complex analytic connections in fibre bundles. Trans.
Amer. Math. Soc., 85:181–207, 1957.
Vikraman Balaji, Indranil Biswas, and Donihakkalu S. Nagaraj. Principal bundles over projective manifolds with parabolic structure over a
divisor. Tohoku Math. J. (2), 53(3):337–367, 2001.
V. Balaji, I. Biswas, and D. S. Nagaraj. Ramified G-bundles as parabolic bundles. J. Ramanujan Math. Soc., 18(2):123–138, 2003.
I. Biswas, T.L. Goméz, and V. Muñoz. Torelli theorem for the
moduli space of framed bundles. arXiv, math.AG, August 2008,
arXiv:0808.1996.
U. N. Bhosle. Parabolic vector bundles on curves. Ark. Mat., 27(1):15–
22, 1989.
Olivier Biquard. Fibrés paraboliques stables et connexions singulières
plates. Bull. Soc. Math. France, 119(2):231–257, 1991.
Arnaud Beauville, M. S. Narasimhan, and S. Ramanan. Spectral curves
and the generalised theta divisor. J. Reine Angew. Math., 398:169–179,
1989.
Francesco Bottacin. Symplectic geometry on moduli spaces of stable
pairs. Ann. Sci. École Norm. Sup. (4), 28(4):391–433, 1995.
Usha Bhosle and A. Ramanathan. Moduli of parabolic G-bundles on
curves. Math. Z., 202(2):161–180, 1989.
[HJS05] the construction given further fixes the determinant for the sheaves,
but the construction goes through without imposing this as well.
30
[BY96]
MARINA LOGARES AND JOHAN MARTENS
Hans U. Boden and Kôji Yokogawa. Moduli spaces of parabolic Higgs
bundles and parabolic K(D) pairs over smooth curves. I. Internat. J.
Math., 7(5):573–598, 1996.
[BY99]
Hans U. Boden and Kôji Yokogawa. Rationality of moduli spaces of
parabolic bundles. J. London Math. Soc. (2), 59(2):461–478, 1999.
[CdSW99] Ana Cannas da Silva and Alan Weinstein. Geometric models for noncommutative algebras, volume 10 of Berkeley Mathematics Lecture
Notes. American Mathematical Society, Providence, RI, 1999.
[CDW87] A. Coste, P. Dazord, and A. Weinstein. Groupoı̈des symplectiques.
In Publications du Département de Mathématiques. Nouvelle Série. A,
Vol. 2, volume 87 of Publ. Dép. Math. Nouvelle Sér. A, pages i–ii, 1–62.
Univ. Claude-Bernard, Lyon, 1987.
[CF03]
Marius Crainic and Rui Loja Fernandes. Integrability of Lie brackets.
Ann. of Math. (2), 157(2):575–620, 2003.
[CG97]
Neil Chriss and Victor Ginzburg. Representation theory and complex
geometry. Birkhäuser Boston Inc., Boston, MA, 1997.
[Cou90]
Theodore James Courant. Dirac manifolds. Trans. Amer. Math. Soc.,
319(2):631–661, 1990.
[Deu73]
Max Deuring. Lectures on the theory of algebraic functions of one variable. Lecture Notes in Mathematics, Vol. 314. Springer-Verlag, Berlin,
1973.
[DK07]
Brent Doran and Frances Kirwan. Towards non-reductive geometric
invariant theory. Pure Appl. Math. Q., 3(1):61–105, 2007.
[DM96]
Ron Donagi and Eyal Markman. Spectral covers, algebraically completely integrable, Hamiltonian systems, and moduli of bundles. In Integrable systems and quantum groups (Montecatini Terme, 1993), volume 1620 of Lecture Notes in Math., pages 1–119. Springer, Berlin,
1996.
[Don03]
Ron Y. Donagi. Geometry and integrability. In Geometry and integrability, volume 295 of London Math. Soc. Lecture Note Ser., pages 21–59.
Cambridge Univ. Press, Cambridge, 2003.
[Fal93]
Gerd Faltings. Stable G-bundles and projective connections. J. Algebraic Geom., 2(3):507–568, 1993.
[GPGM07] O. Garcı́a-Prada, P. B. Gothen, and V. Muñoz. Betti numbers of the
moduli space of rank 3 parabolic Higgs bundles. Mem. Amer. Math.
Soc., 187(879):viii+80, 2007.
[EGA]
A. Grothendieck. Éléments de géométrie algébrique. IV. Étude locale
des schémas et des morphismes de schémas. II. Inst. Hautes Études Sci.
Publ. Math., (24):231, 1965.
[Har77]
Robin Hartshorne. Algebraic geometry. Springer-Verlag, New York,
1977. Graduate Texts in Mathematics, No. 52.
[Hit87a]
N. J. Hitchin. The self-duality equations on a Riemann surface. Proc.
London Math. Soc. (3), 55(1):59–126, 1987.
[Hit87b]
Nigel Hitchin. Stable bundles and integrable systems. Duke Math. J.,
54(1):91–114, 1987.
[HJS05]
Jacques Hurtubise, Lisa Jeffrey, and Reyer Sjamaar. Moduli of framed
parabolic sheaves. Ann. Global Anal. Geom., 28(4):351–370, 2005.
MODULI OF PARABOLIC HIGGS BUNDLES AND ATIYAH ALGEBROIDS 31
[HJS06]
[HL95]
[HL97]
[Hu06]
[Kir08a]
[Kir08b]
[Kon93]
[LN04]
[Lun73]
[Mac05]
[Mar94]
[GIT]
[MS80]
[Nit91]
[NS65]
[Ser58]
[Ses82]
[Sim90]
[Tha02]
Jacques Hurtubise, Lisa Jeffrey, and Reyer Sjamaar. Group-valued implosion and parabolic structures. Amer. J. Math., 128(1):167–214, 2006.
D. Huybrechts and M. Lehn. Stable pairs on curves and surfaces. J.
Algebraic Geom., 4(1):67–104, 1995.
Daniel Huybrechts and Manfred Lehn. The geometry of moduli spaces of
sheaves. Aspects of Mathematics, E31. Friedr. Vieweg & Sohn, Braunschweig, 1997.
Yi Hu. Quotients by reductive group, Borel subgroup, unipotent group
and maximal torus. Pure Appl. Math. Q., 2(4):1131–1147, 2006.
Frances C Kirwan. Quotients by non-reductive algebraic group actions.
arXiv, math.AG, Jan 2008, arXiv:0801.4607v3.
Frances C. Kirwan. Symplectic implosion and non-reductive quotients.
Preprint, 2008.
Hiroshi Konno. Construction of the moduli space of stable parabolic
Higgs bundles on a Riemann surface. J. Math. Soc. Japan, 45(2):253–
276, 1993.
G. Laumon and B.C. Ngô. Le lemme fondamental pour les groupes
unitaires. arXiv, April 2004, arXiv:math/0404454.
Domingo Luna. Slices étales. In Sur les groupes algébriques, pages 81–
105. Bull. Soc. Math. France, Paris, Mémoire 33. Soc. Math. France,
Paris, 1973.
Kirill C. H. Mackenzie. General theory of Lie groupoids and Lie algebroids, volume 213 of London Mathematical Society Lecture Note Series.
Cambridge University Press, Cambridge, 2005.
Eyal Markman. Spectral curves and integrable systems. Compositio
Math., 93(3):255–290, 1994.
D. Mumford, J. Fogarty, and F. Kirwan. Geometric invariant theory, volume 34 of Ergebnisse der Mathematik und ihrer Grenzgebiete
(2) [Results in Mathematics and Related Areas (2)]. Springer-Verlag,
Berlin, third edition, 1994.
V. B. Mehta and C. S. Seshadri. Moduli of vector bundles on curves
with parabolic structures. Math. Ann., 248(3):205–239, 1980.
Nitin Nitsure. Moduli space of semistable pairs on a curve. Proc. London
Math. Soc. (3), 62(2):275–300, 1991.
M. S. Narasimhan and C. S. Seshadri. Stable and unitary vector bundles
on a compact Riemann surface. Ann. of Math. (2), 82:540–567, 1965.
Jean-Pierre Serre. Les espaces fibrés algébriques. In Séminaire C.
Chevalley; 2e année: 1958. Anneaux de Chow et applications, pages
iii+134 pp. (mimeographed). Secrétariat mathématique, 11 rue Pierre
Curie, Paris, 1958.
C. S. Seshadri. Fibrés vectoriels sur les courbes algébriques, volume 96 of
Astérisque. Société Mathématique de France, Paris, 1982. Notes written
by J.-M. Drezet from a course at the École Normale Supérieure, June
1980.
Carlos T. Simpson. Harmonic bundles on noncompact curves. J. Amer.
Math. Soc., 3(3):713–770, 1990.
Michael Thaddeus. Variation of moduli of parabolic Higgs bundles. J.
Reine Angew. Math., 547:1–14, 2002.
32
[TW03]
[Yok93]
[Yun08]
MARINA LOGARES AND JOHAN MARTENS
C. Teleman and C. Woodward. Parabolic bundles, products of conjugacy classes and Gromov-Witten invariants. Ann. Inst. Fourier (Grenoble), 53(3):713–748, 2003.
Kôji Yokogawa. Compactification of moduli of parabolic sheaves and
moduli of parabolic Higgs sheaves. J. Math. Kyoto Univ., 33(2):451–
504, 1993.
Zhiwei Yun. Towards a global Springer theory I: the affine Weyl group
action. arXiv, math.AG, October 2008, arXiv:0810.2146.
Departamento de Matemática Pura, Faculdade de Ciências, Universidade do Porto, Rua do Campo Alegre, 687, 4169-007 Porto, Portugal
E-mail address: [email protected]
Department of Mathematics, University of Toronto, 40 St. George
Street, Toronto Ontario M5S 2E4, Canada
E-mail address: [email protected]